Astor Press est. MMXXVI Download · Mac
Astor Press · Working Paper Nº 1 MMXXVI · In effect

The Method.

Why the volumes are built this way — the evidence, set in order.

Subject · The engineering of understanding Articles · Five Vow · One References · Thirteen
Abstract

The science here is not new. It is merely inconvenient — nearly every result below says the opposite of what an engagement metric wants to hear. What follows is what the press is built on, with its receipts: five articles, a vow, and the wider ledger.

Keywords retrieval practice · generation effect · desirable difficulty · refutation · multimedia principle

“Learning is not supposed to be fun… the primary feeling should be that of effort. You want the mental equivalent of sweating.”

Andrej Karpathy · on the “shortification” of learning · 2024
The Articles five results, with receipts
I
Article I

The feeling of learning lies.

Harvard ran the experiment: physics students made to work through problems learned nearly half a standard deviation more than classmates given a fluent lecture — and walked out convinced they had learned less. Fluency flatters. The press optimizes for what you hold a month later, not the feeling at the door.[1]

II
Article II

The book asks before it tells.

Where a volume pauses to ask what you expect before it shows you, that is not decoration — committing to a guess is one of the most reliable effects in learning science, measured across sixty-nine controlled comparisons. Guess. Then read on.[2]

III
Article III

The closing questions come unanswered.

A week after studying, students who tested themselves held 56 percent of the material; students who reread held 42. So every volume ends with problems, and the margin will coach you toward an answer it will not hand over.[3]

IV
Article IV

The wrong idea is named first.

A misconception outlives polite silence. Volumes name the belief you likely arrived with, show exactly where it breaks, and set the correct idea in the crater — refutation beats clean exposition across forty-four controlled comparisons.[4]

V
Article V

A plate must earn its page.

Prose that arrives with the right picture outteaches prose alone by one of the widest margins measured in learning science — so every figure is engraved for the sentence beside it, and a figure with no work to do is cut.[5]

The Vow

No streaks. No badges.
No leaderboards. No autoplay.

When a semester-long course was run twice — once plain, once dressed in badges and a leaderboard — the gamified section finished the semester less motivated, and their final-exam scores followed. Engagement mechanics are a loan taken out against your own curiosity. The press does not borrow.[6]

Hanus & Fox, Computers & Education, 2015.
References & the wider ledger read further, if you like
  1. [1]Deslauriers, McCarty, Miller, Callaghan & Kestin · PNAS · 2019Active instruction taught more; fluent lecture felt like more.
  2. [2]Bisra, Liu, Nesbit, Salimi & Winne · Educ. Psych. Review · 2018Eliciting a guess before instruction: g = 0.55 across 69 comparisons.
  3. [3]Roediger & Karpicke · Psychological Science · 2006Self-testing held 56% at one week; rereading held 42%.
  4. [4]Schroeder & Kucera · Educ. Psych. Review · 2022Refutation texts beat clean exposition, g = 0.41 across 44 comparisons.
  5. [5]Mayer · Cambridge Handbook of Multimedia LearningWords with the right pictures: median d ≈ 1.4 over words alone.
  6. [6]Hanus & Fox · Computers & Education · 2015Badges and leaderboards lowered motivation — and final exams followed.
  7. [7]Karpicke & Blunt · Science · 2011Retrieval practice beat elaborate concept-mapping at one week (d = 1.50). Students predicted the opposite.
  8. [8]Cepeda, Pashler, Vul, Wixted & Rohrer · 2006Spacing lifts recall from 37 to 47 percent — 254 studies, 14,811 participants.
  9. [9]Adesope, Trevisan & Sundararajan · 2017Practice testing beats restudying across 272 controlled effects (g = 0.51).
  10. [10]Bertsch, Pesta, Wiscott & McDaniel · 2007What you generate yourself, you keep — d = 0.40 across 86 studies.
  11. [11]Dunlosky, Rawson, Marsh, Nathan & Willingham · 2013The habits that feel most natural — rereading, highlighting — rate “low utility.”
  12. [12]Szpunar, Khan & Schacter · 2013Interpolated questions cut mind-wandering from 40% to 19% — and lifted scores from 68 to 90.
  13. [13]Carpenter, Wilford, Kornell & Mullaney · 2013A fluent lecturer inflates the feeling of learning — and adds nothing to the fact of it.

This is the method. Every volume is bound to it.

Free on the Mac App Store
← Return to the press